Previsione sottoscrizione deposito

Nello svolgimento del seguente progetto si tenta di prevedere, mediante l'utilizzo di algoritmi di Machine Learning, l'adesione ad un conto di deposito offerto dalla banca. Viene utilizzato un dataset contenente le informazioni relative ad alcuni clienti della banca.

import tensorflow as tf
import numpy as np
import os
import random
import pandas as pd
import seaborn as sns
from datetime import datetime
import matplotlib.pyplot as plt
from sklearn.preprocessing import MinMaxScaler

from google.colab import drive
drive.mount("/content/drive")

Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount("/content/drive", force_remount=True).

data = pd.read_csv("/content/drive/MyDrive/Colab Notebooks/Classificazione Credito/dati_new1.csv")

Una prima visualizzazione del dataset di partenza.

data.head()

	age	job	marital	education	default	balance	housing	loan	day	month	duration	campaign	pdays	previous	term_deposit
0	42	entrepreneur	divorced	tertiary	yes	2	yes	no	5	may	380	1	-1	0	no
1	43	technician	single	secondary	no	593	yes	no	5	may	55	1	-1	0	no
2	57	services	married	secondary	no	162	yes	no	5	may	174	1	-1	0	no
3	45	admin.	single	unknown	no	13	yes	no	5	may	98	1	-1	0	no
4	60	retired	married	primary	no	60	yes	no	5	may	219	1	-1	0	no

Viene visualizzata la variabile di riferimento, ovvero l'avvenuta adesione ad un conto di deposito sottoscritto dalla banca.

import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline
plt.style.use("ggplot")

sns.countplot(y=data.term_deposit ,data=data)
plt.xlabel("conteggio per classe")
plt.ylabel("classi")
plt.show()

data["term_deposit"].value_counts()

no     8678
yes    3961
Name: term_deposit, dtype: int64

# 54% no 46% si

(data["term_deposit"].value_counts())/len(data)

no     0.686605
yes    0.313395
Name: term_deposit, dtype: float64

Informazioni relative alla tipologia delle variabili, le colonne, il conteggio dei valori nulli e l'utilizzo di memoria.

data.info(verbose=True)

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 12639 entries, 0 to 12638
Data columns (total 15 columns):
 #   Column        Non-Null Count  Dtype 
---  ------        --------------  ----- 
 0   age           12639 non-null  int64 
 1   job           12639 non-null  object
 2   marital       12639 non-null  object
 3   education     12639 non-null  object
 4   default       12639 non-null  object
 5   balance       12639 non-null  int64 
 6   housing       12639 non-null  object
 7   loan          12639 non-null  object
 8   day           12639 non-null  int64 
 9   month         12639 non-null  object
 10  duration      12639 non-null  int64 
 11  campaign      12639 non-null  int64 
 12  pdays         12639 non-null  int64 
 13  previous      12639 non-null  int64 
 14  term_deposit  12639 non-null  object
dtypes: int64(7), object(8)
memory usage: 1.4+ MB

Statistiche descrittive basilare del dataset.

data.describe(include="all")

	age	job	marital	education	default	balance	housing	loan	day	month	duration	campaign	pdays	previous	term_deposit
count	12639.000000	12639	12639	12639	12639	12639.000000	12639	12639	12639.000000	12639	12639.000000	12639.000000	12639.000000	12639.000000	12639
unique	NaN	12	3	4	2	NaN	2	2	NaN	12	NaN	NaN	NaN	NaN	2
top	NaN	management	married	secondary	no	NaN	yes	no	NaN	may	NaN	NaN	NaN	NaN	no
freq	NaN	2761	7399	6332	12419	NaN	6803	10674	NaN	3702	NaN	NaN	NaN	NaN	8678
mean	40.939077	NaN	NaN	NaN	NaN	1425.023261	NaN	NaN	15.715088	NaN	336.726719	2.702745	35.591502	0.507319	NaN
std	10.840946	NaN	NaN	NaN	NaN	3014.382883	NaN	NaN	8.323035	NaN	336.118980	2.962794	91.862379	1.785123	NaN
min	18.000000	NaN	NaN	NaN	NaN	-6847.000000	NaN	NaN	1.000000	NaN	3.000000	1.000000	-1.000000	0.000000	NaN
25%	32.000000	NaN	NaN	NaN	NaN	86.000000	NaN	NaN	8.000000	NaN	119.000000	1.000000	-1.000000	0.000000	NaN
50%	39.000000	NaN	NaN	NaN	NaN	478.000000	NaN	NaN	16.000000	NaN	219.000000	2.000000	-1.000000	0.000000	NaN
75%	48.000000	NaN	NaN	NaN	NaN	1516.500000	NaN	NaN	21.000000	NaN	436.000000	3.000000	-1.000000	0.000000	NaN
max	95.000000	NaN	NaN	NaN	NaN	81204.000000	NaN	NaN	31.000000	NaN	3881.000000	43.000000	520.000000	58.000000	NaN

data.isnull().sum()

age             0
job             0
marital         0
education       0
default         0
balance         0
housing         0
loan            0
day             0
month           0
duration        0
campaign        0
pdays           0
previous        0
term_deposit    0
dtype: int64

data.columns

Index(['age', 'job', 'marital', 'education', 'default', 'balance', 'housing',
       'loan', 'day', 'month', 'duration', 'campaign', 'pdays', 'previous',
       'term_deposit'],
      dtype='object')

Viene stabilito quali siano le covariate numeriche e quali siano le covariate categoriche.

data.dtypes

age              int64
job             object
marital         object
education       object
default         object
balance          int64
housing         object
loan            object
day              int64
month           object
duration         int64
campaign         int64
pdays            int64
previous         int64
term_deposit    object
dtype: object

Verifica delle modalità delle covariate categoriche.

for column in data.columns:
    if data[column].dtype == "object":
        print(column.upper())
        print('=====================================')
        print(data[column].unique() ,"\n\n\n")

JOB
=====================================
['entrepreneur' 'technician' 'services' 'admin.' 'retired' 'blue-collar'
 'management' 'self-employed' 'unemployed' 'student' 'unknown' 'housemaid'] 



MARITAL
=====================================
['divorced' 'single' 'married'] 



EDUCATION
=====================================
['tertiary' 'secondary' 'unknown' 'primary'] 



DEFAULT
=====================================
['yes' 'no'] 



HOUSING
=====================================
['yes' 'no'] 



LOAN
=====================================
['no' 'yes'] 



MONTH
=====================================
['may' 'jun' 'jul' 'aug' 'oct' 'nov' 'dec' 'jan' 'feb' 'mar' 'apr' 'sep'] 



TERM_DEPOSIT
=====================================
['no' 'yes'] 

Ricerca di tutte le colonne con valori nulli.

def unknown(data_frame):
    columns = data_frame.columns

    presence_list = []
    for i in columns:
        if "unknown" in data_frame[i].unique():
            presence_list.append(i)
    return presence_list

unknowns = unknown(data)
unknowns

<ipython-input-16-6d709e058206>:6: FutureWarning: elementwise comparison failed; returning scalar instead, but in the future will perform elementwise comparison
  if "unknown" in data_frame[i].unique():

['job', 'education']

for i in unknowns:
    precentage = len(data[data[i] == "unknown"]) /len(data[i]) * 100
    print(i.upper() , f"\n=========================================\n{str(precentage)}" + "%" "\n\n\n" )

JOB 
=========================================
0.6171374317588417%



EDUCATION 
=========================================
3.766120737400111%

Creazione delle funzioni per osservare la distribuzione delle "features".

# creazione di una funzione per visualizzare la distribuzione di tutte le features
def plot_distributions(df):
    plt.figure(figsize=(28,40))
    b = 0
    for i in df.columns:
        b+=1
        plt.subplot(6,6,b)
        plt.hist(df[i])
        plt.title(i)

# creazione di una funzione per visualizzare la distribuzione di una sola feature
def plot_idividual( column_name,df= data):
    plt.figure(figsize=(15,10))
    plt.hist(df[column_name])
    plt.title(column_name)

# Grafico di distribuzione delle features
plot_distributions(data)

df_categorical=data.select_dtypes(include=['object'])
df_categorical.head()

	job	marital	education	default	housing	loan	month	term_deposit
0	entrepreneur	divorced	tertiary	yes	yes	no	may	no
1	technician	single	secondary	no	yes	no	may	no
2	services	married	secondary	no	yes	no	may	no
3	admin.	single	unknown	no	yes	no	may	no
4	retired	married	primary	no	yes	no	may	no

df_numerical=data.select_dtypes(include=['int','float'])
df_numerical.head()

	age	balance	day	duration	campaign	pdays	previous
0	42	2	5	380	1	-1	0
1	43	593	5	55	1	-1	0
2	57	162	5	174	1	-1	0
3	45	13	5	98	1	-1	0
4	60	60	5	219	1	-1	0

Variabili categoriche

df_0=df_categorical[data["term_deposit"]=="no"] # records con target==no
df_1=df_categorical[data["term_deposit"]=="yes"] # records con target==yes
for x in df_categorical.columns :
    plt.figure()
    plt.hist([df_1[x],df_0[x]],density=True )
    plt.title(x)
    plt.show()

df_categorical.columns

Index(['job', 'marital', 'education', 'default', 'housing', 'loan', 'month',
       'term_deposit'],
      dtype='object')

Trasformazione delle variabili categoriche in dummies.

dummies = pd.get_dummies(df_categorical[["job", "marital", "education", "default", "housing", "loan", "month", "term_deposit"]],drop_first=True)

dummies.tail()

	job_blue-collar	job_entrepreneur	job_housemaid	job_management	job_retired	job_self-employed	job_services	job_student	job_technician	job_unemployed	...	month_feb	month_jan	month_jul	month_jun	month_mar	month_may	month_nov	month_oct	month_sep	term_deposit_yes
12634	0	0	0	0	0	1	0	0	0	0	...	0	1	0	0	0	0	0	0	0	1
12635	0	0	0	1	0	0	0	0	0	0	...	0	1	0	0	0	0	0	0	0	1
12636	0	0	0	0	0	0	0	1	0	0	...	0	1	0	0	0	0	0	0	0	1
12637	0	0	0	0	1	0	0	0	0	0	...	0	1	0	0	0	0	0	0	0	1
12638	0	0	0	0	0	0	0	1	0	0	...	0	1	0	0	0	0	0	0	0	1

5 rows × 31 columns

dummies.columns

Index(['job_blue-collar', 'job_entrepreneur', 'job_housemaid',
       'job_management', 'job_retired', 'job_self-employed', 'job_services',
       'job_student', 'job_technician', 'job_unemployed', 'job_unknown',
       'marital_married', 'marital_single', 'education_secondary',
       'education_tertiary', 'education_unknown', 'default_yes', 'housing_yes',
       'loan_yes', 'month_aug', 'month_dec', 'month_feb', 'month_jan',
       'month_jul', 'month_jun', 'month_mar', 'month_may', 'month_nov',
       'month_oct', 'month_sep', 'term_deposit_yes'],
      dtype='object')

Variabili Numeriche

df_numerical.hist()

array([[<Axes: title={'center': 'age'}>,
        <Axes: title={'center': 'balance'}>,
        <Axes: title={'center': 'day'}>],
       [<Axes: title={'center': 'duration'}>,
        <Axes: title={'center': 'campaign'}>,
        <Axes: title={'center': 'pdays'}>],
       [<Axes: title={'center': 'previous'}>, <Axes: >, <Axes: >]],
      dtype=object)

df_numerical.previous

0        0
1        0
2        0
3        0
4        0
        ..
12634    0
12635    2
12636    1
12637    0
12638    0
Name: previous, Length: 12639, dtype: int64

Trasformazione funzionale delle covariate numeriche.

import math
#trasformazione delle covariate

df_numerical["target"]=dummies["term_deposit_yes"]
df_numerical["logcampaign"]=df_numerical["campaign"].apply(math.log)
df_numerical["logduration"]=df_numerical["duration"].apply(lambda x: math.log(x+1))
df_numerical["logprevious"]=df_numerical["previous"].apply(lambda x: math.log(x+1))

df_numerical.tail()

	age	balance	day	duration	campaign	pdays	previous	target	logcampaign	logduration	logprevious
12634	28	4579	12	409	2	-1	0	1	0.693147	6.016157	0.000000
12635	49	1093	12	243	2	91	2	1	0.693147	5.497168	1.098612
12636	21	2488	12	661	2	92	1	1	0.693147	6.495266	0.693147
12637	87	2190	12	512	2	-1	0	1	0.693147	6.240276	0.000000
12638	22	254	13	143	2	-1	0	1	0.693147	4.969813	0.000000

df_numerical.hist()

array([[<Axes: title={'center': 'age'}>,
        <Axes: title={'center': 'balance'}>,
        <Axes: title={'center': 'day'}>],
       [<Axes: title={'center': 'duration'}>,
        <Axes: title={'center': 'campaign'}>,
        <Axes: title={'center': 'pdays'}>],
       [<Axes: title={'center': 'previous'}>,
        <Axes: title={'center': 'target'}>,
        <Axes: title={'center': 'logcampaign'}>],
       [<Axes: title={'center': 'logduration'}>,
        <Axes: title={'center': 'logprevious'}>, <Axes: >]], dtype=object)

Grafico che mette in relazione le variabili numeriche con riferimento alla variabile target, ovvero l'adesione o meno al conto di deposito (rispettivamente in rosso e in blu).

sns.pairplot(df_numerical[["age",	"balance",	"day",	"duration",	"campaign",	"pdays",	"previous",	"logcampaign",	"logduration",	"logprevious", "target"]], hue="target")

<seaborn.axisgrid.PairGrid at 0x7c5332634b50>

Trasformazione in dati scalati.

from sklearn.preprocessing import StandardScaler

scaler = StandardScaler().fit(df_numerical)
scaled_df = pd.DataFrame(scaler.transform(df_numerical))
scaled_df.columns = df_numerical.columns

scaled_df.tail()

	age	balance	day	duration	campaign	pdays	previous	target	logcampaign	logduration	logprevious
12634	-1.193585	1.046351	-0.446380	0.215031	-0.2372	-0.398345	-0.284204	1.480156	0.003036	0.639552	-0.409425
12635	0.743592	-0.110151	-0.446380	-0.278861	-0.2372	0.603192	0.836211	1.480156	0.003036	0.106618	1.777717
12636	-1.839310	0.352649	-0.446380	0.964796	-0.2372	0.614079	0.276004	1.480156	0.003036	1.131533	0.970508
12637	4.248960	0.253786	-0.446380	0.521483	-0.2372	-0.398345	-0.284204	1.480156	0.003036	0.869692	-0.409425
12638	-1.747064	-0.388494	-0.326227	-0.576386	-0.2372	-0.398345	-0.284204	1.480156	0.003036	-0.434906	-0.409425

scaled_df.boxplot()

<Axes: >

scaled_df.tail()

	age	balance	day	duration	campaign	pdays	previous	target	logcampaign	logduration	logprevious
12634	-1.193585	1.046351	-0.446380	0.215031	-0.2372	-0.398345	-0.284204	1.480156	0.003036	0.639552	-0.409425
12635	0.743592	-0.110151	-0.446380	-0.278861	-0.2372	0.603192	0.836211	1.480156	0.003036	0.106618	1.777717
12636	-1.839310	0.352649	-0.446380	0.964796	-0.2372	0.614079	0.276004	1.480156	0.003036	1.131533	0.970508
12637	4.248960	0.253786	-0.446380	0.521483	-0.2372	-0.398345	-0.284204	1.480156	0.003036	0.869692	-0.409425
12638	-1.747064	-0.388494	-0.326227	-0.576386	-0.2372	-0.398345	-0.284204	1.480156	0.003036	-0.434906	-0.409425

X_numerical=scaled_df[['age',	'balance',	'day',	'duration',	'campaign',	'pdays',	'previous',	'logcampaign',	'logduration',	'logprevious']]

X_dummies= dummies.drop('term_deposit_yes', axis=1)

print(X_dummies.shape)
print(X_numerical.shape)

X_dummies.head()

(12639, 30)
(12639, 10)

	job_blue-collar	job_entrepreneur	job_housemaid	job_management	job_retired	job_self-employed	job_services	job_student	job_technician	job_unemployed	...	month_dec	month_feb	month_jan	month_jul	month_jun	month_mar	month_may	month_nov	month_oct	month_sep
0	0	1	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	1	0	0	0
1	0	0	0	0	0	0	0	0	1	0	...	0	0	0	0	0	0	1	0	0	0
2	0	0	0	0	0	0	1	0	0	0	...	0	0	0	0	0	0	1	0	0	0
3	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	1	0	0	0
4	0	0	0	0	1	0	0	0	0	0	...	0	0	0	0	0	0	1	0	0	0

5 rows × 30 columns

X=pd.concat([X_dummies,X_numerical], axis = 1)

print(X.shape)

X.tail()

(12639, 40)

	job_blue-collar	job_entrepreneur	job_housemaid	job_management	job_retired	job_self-employed	job_services	job_student	job_technician	job_unemployed	...	age	balance	day	duration	campaign	pdays	previous	logcampaign	logduration	logprevious
12634	0	0	0	0	0	1	0	0	0	0	...	-1.193585	1.046351	-0.446380	0.215031	-0.2372	-0.398345	-0.284204	0.003036	0.639552	-0.409425
12635	0	0	0	1	0	0	0	0	0	0	...	0.743592	-0.110151	-0.446380	-0.278861	-0.2372	0.603192	0.836211	0.003036	0.106618	1.777717
12636	0	0	0	0	0	0	0	1	0	0	...	-1.839310	0.352649	-0.446380	0.964796	-0.2372	0.614079	0.276004	0.003036	1.131533	0.970508
12637	0	0	0	0	1	0	0	0	0	0	...	4.248960	0.253786	-0.446380	0.521483	-0.2372	-0.398345	-0.284204	0.003036	0.869692	-0.409425
12638	0	0	0	0	0	0	0	1	0	0	...	-1.747064	-0.388494	-0.326227	-0.576386	-0.2372	-0.398345	-0.284204	0.003036	-0.434906	-0.409425

5 rows × 40 columns

Separazione training set e test set

y=df_numerical['target']
y.shape

(12639,)

from sklearn.model_selection import train_test_split

#SEPARAZIONE DEI DATI IN TRAIN SET E TEST SET
X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                    test_size =0.30,
                                                    stratify=y, #preservo la proporzione dei dati
                                                    random_state= 123) #fisso un seed per replicare

print(X_train.shape, X_test.shape)

(8847, 40) (3792, 40)

Modelli

Importo i modelli che utilizzerò per lo sviluppo dell'algoritmo.

from sklearn.model_selection import GridSearchCV
from sklearn.metrics import confusion_matrix
from sklearn.metrics import classification_report
from sklearn import metrics
from sklearn.metrics import f1_score
from sklearn.neural_network import MLPClassifier

# KNN
from sklearn.neighbors import KNeighborsClassifier
classifier = KNeighborsClassifier()
parameters = {"n_neighbors":np.arange(10,100,1000)}

def hyperp_search(classifier, parameters):
    gs = GridSearchCV(classifier, parameters, cv=3, scoring = "f1", verbose=0, n_jobs=-1)
    gs = gs.fit(X_train, y_train)
    print("f1_train: %f con %s" % (gs.best_score_, gs.best_params_))

    best_model = gs.best_estimator_
    y_pred = best_model.predict(X_test)

    print("f1_test: ", f1_score(y_test, y_pred))
    print(confusion_matrix(y_test, y_pred))
    print(classification_report(y_test, y_pred))

hyperp_search(classifier,parameters)

f1_train: 0.643011 con {'n_neighbors': 10}
f1_test:  0.6857697911607576
[[2439  165]
 [ 482  706]]
              precision    recall  f1-score   support

           0       0.83      0.94      0.88      2604
           1       0.81      0.59      0.69      1188

    accuracy                           0.83      3792
   macro avg       0.82      0.77      0.78      3792
weighted avg       0.83      0.83      0.82      3792

model_knn = KNeighborsClassifier(n_neighbors=10)

def roc(model,X_train,y_train,X_test,y_test):
    model.fit(X_train, y_train)
    y_pred = model.predict(X_test)

    y_probs = model.predict_proba(X_test) #predict_proba restituisce le probabilità per il target (0 e 1 in questo caso)

    fpr, tpr, thresholds1=metrics.roc_curve(y_test,  y_probs[:,1])

    import matplotlib.pyplot as plt
    plt.plot(fpr, tpr, label="ROC")
    plt.plot([0, 1], [0, 1], color="darkblue", linestyle="--")
    plt.xlabel("False Positive Rate")
    plt.ylabel("True Positive Rate")
    plt.title("Receiver Operating Characteristic (ROC)")
    plt.legend()
    plt.show()

    auc = metrics.roc_auc_score(y_test, y_probs[:,1])
    print("AUC: %.2f" % auc)
    return (fpr, tpr)

fpr1,tpr1=roc(model_knn,X_train,y_train,X_test,y_test)

AUC: 0.90

#Tree

from sklearn.tree import DecisionTreeClassifier
classifier = DecisionTreeClassifier()
parameters = {"criterion": ["entropy","gini"],
              "max_depth": [4,5,10],
              "min_samples_split": [20],
              "min_samples_leaf": [10]}

hyperp_search(classifier,parameters)

f1_train: 0.705582 con {'criterion': 'entropy', 'max_depth': 10, 'min_samples_leaf': 10, 'min_samples_split': 20}
f1_test:  0.7096204766107679
[[2330  274]
 [ 384  804]]
              precision    recall  f1-score   support

           0       0.86      0.89      0.88      2604
           1       0.75      0.68      0.71      1188

    accuracy                           0.83      3792
   macro avg       0.80      0.79      0.79      3792
weighted avg       0.82      0.83      0.82      3792

model_tree = DecisionTreeClassifier(criterion="entropy", max_depth=5, min_samples_leaf=10, min_samples_split=20)

fpr2,tpr2=roc(model_tree,X_train,y_train,X_test,y_test)

AUC: 0.87

# Naive Bayes
from sklearn.naive_bayes import GaussianNB

model = GaussianNB()

model.fit(X_train, y_train)
y_pred=model.predict(X_test)

from sklearn.metrics import f1_score
print("f1_score: ", f1_score(y_test, y_pred))

print("f1_test: ", f1_score(y_test, y_pred))


print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))

f1_score:  0.5586061246040126
f1_test:  0.5586061246040126
[[2427  177]
 [ 659  529]]
              precision    recall  f1-score   support

           0       0.79      0.93      0.85      2604
           1       0.75      0.45      0.56      1188

    accuracy                           0.78      3792
   macro avg       0.77      0.69      0.71      3792
weighted avg       0.77      0.78      0.76      3792

y_probs = model.predict_proba(X_test) #predict_proba restituisce le probabilità per il target (0 e 1 in questo caso)

fpr3,tpr3=roc(model,X_train,y_train,X_test,y_test)

AUC: 0.84

# Logistic

from sklearn.linear_model import LogisticRegression

classifier = LogisticRegression()
parameters = {"C":[1e-4,1e-3,1e-2,1e-1,1,10], "max_iter":[1000] }


hyperp_search(classifier,parameters)

f1_train: 0.702697 con {'C': 1, 'max_iter': 1000}
f1_test:  0.7315998237108859
[[2353  251]
 [ 358  830]]
              precision    recall  f1-score   support

           0       0.87      0.90      0.89      2604
           1       0.77      0.70      0.73      1188

    accuracy                           0.84      3792
   macro avg       0.82      0.80      0.81      3792
weighted avg       0.84      0.84      0.84      3792

model = LogisticRegression(C=10, max_iter=2000)

fpr4,tpr4=roc(model,X_train,y_train,X_test,y_test)

AUC: 0.91

# Multi-layer Perceptron classifier

from sklearn.neural_network import MLPClassifier

classifier = MLPClassifier()
parameters = {"hidden_layer_sizes":[(10, 5),(100,20,5)],  "max_iter": [2000], "alpha": [0.001,0.1]}

hyperp_search(classifier,parameters)

f1_train: 0.754116 con {'alpha': 0.1, 'hidden_layer_sizes': (10, 5), 'max_iter': 2000}
f1_test:  0.7842816209578387
[[2307  297]
 [ 230  958]]
              precision    recall  f1-score   support

           0       0.91      0.89      0.90      2604
           1       0.76      0.81      0.78      1188

    accuracy                           0.86      3792
   macro avg       0.84      0.85      0.84      3792
weighted avg       0.86      0.86      0.86      3792

model_MLP=MLPClassifier(hidden_layer_sizes=(10,5), alpha=0.1, max_iter=2000)

fpr6,tpr6=roc(model_MLP,X_train,y_train,X_test,y_test)

AUC: 0.92

plt.plot(fpr1, tpr1, label= "KNN")
plt.plot(fpr2, tpr2, label= "Tree")
plt.plot(fpr3, tpr3, label= "NB")
plt.plot(fpr4, tpr4, label= "Logistic")
plt.plot(fpr6, tpr6, label= "NeuralNet")
plt.plot([0, 1], [0, 1], color="darkblue", linestyle="--")
plt.xlabel("False Positive Rate")
plt.ylabel("True Positive Rate")
plt.title("Receiver Operating Characteristic (ROC)")
plt.legend()
plt.show()

Conclusione

I risultati migliori sono forniti dal classificatore "Multi-layer Perceptron". f1 score del training set è pari a 75.4% e f1 score del test set è pari a 78.4%. La precisione e il recall nella classificazione degli esiti positivi sono rispettivamente 76% e 81%. La previsione di apertura di nuovi conti di deposito presso la banca avviene con una precisione dell' 86.1%.